Speech & audio

Score a text-to-speech provider on your data

Build a text-to-speech dataset from your own sentences with one or more providers, then let reviewers rate each clip for naturalness and accent, side by side. Whisper transcribes every clip back so the word error rate comes for free.

CPU containers for hosted providers; a GPU container for Kokoro on the cluster and for the Whisper check.

Open Datasets Datasets docs

How it works

1Import: sentences from an existing dataset, a volume, a spreadsheet or a URL list.
2Synthesize: Kokoro on the cluster, or ElevenLabs, Azure Neural or Google TTS with your keys as workspace secrets; the clip is the example, so a voice is required.
3Verify: Whisper transcribes each clip and the word error rate is stored with it.
4Rate: reviewers listen and score clip quality in the annotation workspace; export as TSV to compare providers.

The code

lab/ingest

cd lab/ingest
python app.py --name nhc-voice --kind tts --source dataset --dataset nhc-sept \
  --provider azure --voice ar-SA-HamedNeural

run it

# or: Datasets → New dataset → kind tts → provider and voice

Where to go next

Open Datasetskind tts Datasets docsproviders per kind Sharing docsreviewer links and the NDA gate

Transcribe calls in batches with Whisper

Drop recordings on a volume, get diarised transcripts as a dataset, with a person correcting the drafts.

Teach a call agent with human feedback

Play the seven missions: calls in, hot spots, duels for 10 XP a pick, level up, boss fight, crown the champion, connect it to real customers.

Share files between jobs on a volume

Persistent folders mounted into every container; browse, upload and download from the dashboard or the CLI.